CNV analysis using sequencing data?

[foxyg]

Hi Guys,

I have exome sequencing data, I am interested in knowing the CNV information.

Does anyone know the best tools for CNV analysis on sequencing data?

Thanks

[csoong]

I would think with exome data there might be too big a bias due to selection and pcr. A crude CNV info could be extracted though

[krobison]

It has certainly been claimed that you can still get a CNV signal (similarly, selection of RNA is claimed to retain expression information); still, I would think you'd want a lot of samples enriched with the same strategy.

In any case, there are a bunch of tools in this space you might try out, though I don't have specific recommendations

[csoong]

K,
You are right about the RNA, but the bait design should be different. Nonetheless, it always bothered me when expression profiling involves amplification steps...
Wasn't there a protocol claiming no amplification procedure for RNA sequencing? I remember it was in Nature last year or a year before

[foxyg]

So is CNV most accurate when it is whole genome sequencing?

[shruti]

Hi,
the tools referred above seem mostly for the whole genome sequencing and not for the targeted capture and sequencing.
Does any one have other suggestions?

[gprakhar]

It really does not matter if its Exome data or WGS, a simple strategy is to ignore the Intronic regions(you would be having the coordinates)

On the other hand if you really want to be very accurate and you sample is something like, Tumor Normal pair then the job is far more easy. As the regions with zero Read count/depth will be common in both the samples so, no false positives that way.

By far I have found CNVnator from the 1000 genomes group to be the best, very high sensitivity and less of false positives across a wide range of CNV size.

[husamia]

I am also interested in doing this as well. What are yo referring to? There are generally paired read approach, read depth approach, split read approach, and sequence assembly approach for general structural variants.

[gprakhar]

Originally Posted by husamia

I am also interested in doing this as well. What are yo referring to? There are generally paired read approach, read depth approach, split read approach, and sequence assembly approach for general structural variants.

For Calculating CNV there are multiple ways as you mentioned, eg. Read Depth or Read Count in a Static Window or Dynamic window, even bins as small as 100 bp have been used.

I have come across two ways of detecting CNV from NGS data.
First is the individual sample approach which you were talking about initially and CNVnator uses. Take a window overlapping/non-overlapping of 'n' bp and based on the read count/read depth inside it, call CNV.

The second approach is a Pair approach(not Paired end data). Where you compare/take ratio of a Sample vs Test. The main advantage is the CNV common in both are ignored(as ration will be roughly ~1). This can be useful if you have Exome only data. As using it with a Exome reference genome will take care of the Intron regions in the sample.

eg. RDXplorer uses Read dept information, CNVseg read count in a static overlapping window, CNVnator can use read count in 100 base pair bins.

So exactly what kind of sample/data do you have?

[jchoo]

Originally Posted by gprakhar

For Calculating CNV there are multiple ways as you mentioned, eg. Read Depth or Read Count in a Static Window or Dynamic window, even bins as small as 100 bp have been used.

I have come across two ways of detecting CNV from NGS data.
First is the individual sample approach which you were talking about initially and CNVnator uses. Take a window overlapping/non-overlapping of 'n' bp and based on the read count/read depth inside it, call CNV.

The second approach is a Pair approach(not Paired end data). Where you compare/take ratio of a Sample vs Test. The main advantage is the CNV common in both are ignored(as ration will be roughly ~1). This can be useful if you have Exome only data. As using it with a Exome reference genome will take care of the Intron regions in the sample.

eg. RDXplorer uses Read dept information, CNVseg read count in a static overlapping window, CNVnator can use read count in 100 base pair bins.

So exactly what kind of sample/data do you have?

beyond now,most of cnv detector are based on the assumption :the distribution of depth are possion disrtibution,and the most important things are the exome sequencing data are not,so...

[gprakhar]

Originally Posted by jchoo

beyond now,most of cnv detector are based on the assumption :the distribution of depth are possion disrtibution,and the most important things are the exome sequencing data are not,so...

I agree most of them are but not all.
Tools like CNVnator, CNVer and some others have other approaches so with a bit of manual intervention you could get the desired results.
Additionally yesterday I read in a post that CLC bio has a CNV detection tool that works well with Exome data.

Regards,
pg

[slp]

Originally Posted by gprakhar

I agree most of them are but not all.
Tools like CNVnator, CNVer and some others have other approaches so with a bit of manual intervention you could get the desired results.
Additionally yesterday I read in a post that CLC bio has a CNV detection tool that works well with Exome data.

Regards,
pg

Is there a suggested bin size for Human samples? I am using 300bp bin size but am getting thousands of calls as output. Do you know how to filter the calls made by CNVnator (e.g. p-value etc).

[gprakhar]

Originally Posted by slp

Is there a suggested bin size for Human samples? I am using 300bp bin size but am getting thousands of calls as output. Do you know how to filter the calls made by CNVnator (e.g. p-value etc).

Well that would depend on the tool first of all, as depending on how the tool works the bin size is decided.
For eg. CNVnator uses a bin size of 100bp, for the high coverage trio data from the 1000 genomes project and 300bp bin size for low coverage data from the project.

But as the author has mentioned in the paper, depending on the data it is variable.

We
 have
 performed a
 sensitivity
 analysis
 to
 determine
 an
 optimal
 bin
 size
 for
 RD
 analysis
 (see
 Supplement).


CNVnator paper, page 4, first line.

Even for our yet to be published CNV detection program we had to do a sensitivity analysis to determine a the optimal bin size.
Finally the bin size is determined depending on the read length and coverage, for our algorithm.

Thanking you,
--
pg

[fadista]

Hi,

I would suggest ExomeCNV: http://bioinformatics.oxfordjournals...rmatics.btr462